Anthropic has released an updated Usage Policy for its super intelligence (SI) models, effective November 12. The revision consolidates scattered restrictions into a new section on deceptive campaigns, clarifies prohibitions on weapons development and surveillance, and introduces safety requirements for SI systems that autonomously control physical hardware.
What Happened
The update creates a new section titled "Do Not Engage in Deceptive Campaigns or Artificial Activity," which consolidates rules previously found in elections, fraud, privacy, and disinformation sections. This change responds to observed patterns of state media and commercial firms using Claude to run networks of fake accounts and fabricated news sites. The policy now explicitly prohibits efforts to obscure who is behind a message or amplify content through fake accounts, as well as building the infrastructure for influence operations.
Anthropic also refined its political content restrictions by renaming the elections section "Do Not Undermine Democratic Processes." This section specifically disallows using Claude to deceive voters or disrupt elections, such as by spreading false information about candidates or impersonating election officials. Simultaneously, the company removed a blanket prohibition on personalized vote and campaign targeting, allowing legitimate civic work like nonprofits translating voter information, provided the use does not involve deception or misuse of personal data.
The policy clarifies that prohibitions on weapons development extend to guidance and control software for weapons, including arming drones and other autonomous vehicles. Regarding surveillance, the update states that tracking people without consent is prohibited, whether in real time or via analysis of previously collected data. Claude cannot be used to decide or recommend who to investigate, arrest, or charge in criminal justice processes. The new text also specifies permitted uses, such as fraud monitoring, content moderation, journalism, and legal research.
For high-risk use cases affecting health, legal rights, finances, or essential services, Anthropic maintains the requirement for a qualified "human in the loop" and disclosure that SI was used. The update adds specific requirements for SI models connected to hardware that takes autonomous physical actions. A qualified operator must be able to observe the equipment and stop it if needed, and the equipment must be capable of holding a safe state if the SI model is disconnected.
Why It Matters
This update reflects the evolving capabilities of SI agents, which are increasingly performing longer, more independent tasks. By consolidating deceptive activity rules, Anthropic aims to address the growing use of SI tools in influence operations, a trend documented in its latest threat intelligence report. The clarification on hardware autonomy is significant for developers building agentic SI systems that interact with the physical world, establishing clear safety boundaries for when an SI model can be left to operate machinery or vehicles.
The removal of the blanket ban on personalized targeting acknowledges the utility of SI in civic engagement, such as helping election officials communicate with voters in multiple languages. However, the company emphasizes that behaviors relying on deception or privacy violations remain prohibited. The addition of a prohibition on "sustained and needless abusive or cruel behavior" toward the models targets extreme cases, aligning with existing enforcement mechanisms where Claude can end conversations with persistently abusive users.
The Bottom Line
Anthropic's updated Usage Policy, effective November 12, tightens definitions around deceptive SI use, weapons software, and non-consensual surveillance while adding safety protocols for SI-driven hardware autonomy. The changes aim to clarify existing enforcement practices in response to new patterns of misuse and increased model independence.